FDT-biotech: composable digital twins for clinical trials projections

The National Science Foundation Division of Mathematical Sciences awarded Arizona State University $799,999 on November 1, 2026, under the Mathematical and Physical Sciences program (CFDA 47.049) to develop composable digital twins and computational frameworks for reconstructing counterfactual clinical-trial outcomes.

The project, which runs through October 31, 2029, addresses a core challenge in clinical research: when medical practice evolves during long-running trials, final results conflate the effects of the experimental intervention with the effects of improved standard-of-care procedures, rendering conclusions inconclusive. The recipient will build patient-specific computational models calibrated to individual clinical measurements, then replay completed trials under alternative treatment scenarios to isolate intervention effects and quantify outcome confidence. The research will evaluate these methods using data from diverse completed clinical trials across multiple diseases and intervention classes affecting millions of Americans. Deliverables include open-source software and regulatory-science tools to enable adoption by device and drug developers, supporting faster, less expensive, and more reliable evaluation of new medical technologies.

Work is performed in Scottsdale, Arizona. Arizona State University holds the award through its Office of Research and Sponsored Projects Administration (Orspa).

Abstract

Medical clinical trials often span many years, during which the standard of care continues to evolve. When physicians adopt new procedures alongside an experimental therapy or device, the final trial results may no longer reveal how much patient benefit arose from the intervention itself and how much resulted from changes in clinical practice. Because patients cannot ethically be denied improved care, this challenge is difficult to avoid and is frequently recognized only after substantial time and resources have been invested. This project will develop computational models that reconstruct how a clinical trial would have unfolded had medical practice remained unchanged, thereby recovering actionable evidence from studies that might otherwise prove inconclusive. The research will create a clinical-trial world model built from patient-specific digital twins?computational representations calibrated to an individual's clinical measurements?that separate the effects of therapies, procedures, and other interventions. The framework will replay completed trials under alternative treatment scenarios and quantify the confidence associated with each outcome. The methods will be evaluated using data from completed clinical trials across diverse diseases and intervention classes, including conditions that affect millions of americans. Open-source software and regulatory-science tools generated by the project will enable broader adoption by device and drug developers, leading to faster, less expensive, and more reliable evaluation of new medical technologies and stronger evidence for treatment decisions. This three-year project will develop composable, uncertainty-quantified computational frameworks for reconstructing counterfactual clinical-trial outcomes from studies affected by evolving clinical practice and human decision making. The work will create a clinical-trial world model that integrates patient-level digital twins, clinician decision models, reference protocols, and cohort partitions into a unified representation of the trial environment, producing treatment-effect estimates together with decomposed sources of uncertainty. Three research thrusts will advance the framework. The first will establish a formal theory of composability for digital twins, identifying the conditions under which models learned from complex clinical data can be decomposed into causally separable intervention channels and combined reliably, while developing validation strategies based on cross-cohort consistency, held-out-site prediction, and physiological plausibility. The second will reconstruct latent clinician decision policies from their effects on treatment allocation and CO-intervention intensity through novel machine-learning and causal-inference methods that recover hidden treatment dynamics and identify patient populations for which reliable counterfactual projections are possible. The third will develop distribution-free uncertainty quantification for cohorts combining observed and twin-projected outcomes, including methods for detecting protocol drift, generating valid confidence intervals under distribution shift, decomposing uncertainty into statistical and model-based sources, and defining effective sample size for counterfactual analyses. The intellectual merit lies in unifying mechanistic modeling, causal inference, and uncertainty quantification into a certifiable theory of digital-twin composability for clinical trials. The framework will be validated on completed multicenter randomized trials spanning diverse diseases and intervention classes and will culminate in an agentic interface capable of answering natural-language regulatory questions with responses traceable to specific model parameters, causal estimates, and uncertainty decompositions. Beyond advancing regulatory science, the project will strengthen education and workforce development in digital twin technologies and computational medicine, helping to prepare the skilled workforce needed across healthcare and other sectors vital to the U.S. Economy. This project is jointly supported by the national institutes of health through the office of data science strategy. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.